Proposal for logging

Logging

The current situation

• There are two logging primitives log and log_err. Both allow a variety of types to be given.

• log_err output is always printed. This is typically used right before a fail, or in case of other serious problems.

• log is a conditional logging statement. Its output is suppressed by default, but can be turned on using the RUST_LOG environment variable.

• RUST_LOG takes a comma-separated list of directives. A directive is, for now, a module name prefix, for example std::io. The 'main' program will live under main, unless an explicit crate name is given in the crate file.

• In the runtime, the LOG and LOG_ERR, (or DLOG_[ERR] if no task pointer is available) serve a similar role as log and log_err. They explicitly set their own 'module'. These modules live under rt, for example ::rt::comm, ::rt::upcall (see rust_log.h for a list).

• Logging (when muffled) is really really cheap in rustc now.

The future

• Logging will be done through a fmt!-like syntax extension, which (somehow) expands to a special form that handles the module-check wiring and such. The syntax extension will handle stringification, so the runtime logger will only deal with strings anymore.

• There will be more log levels: 0 error, 1 warning, 2 note, 3 debug. The default log level will be 1 (show warnings and errors). The RUST_LOG directives will support module=2 syntax to set specific log levels. If no =X is given, the level for that module is set to 3.

• There will be an API (though a library call) to restrict logging to a single domain (or maybe task). (Maybe also a set of domains, if those become referenceable somehow.)

• There will also be an API that allows you to pass in your on RUST_LOG-style string to reset the per-module logging levels at runtime.